TDT: the purpose of mean substraction in RSA

Hello everyone,

I have a question regarding the scaling method in Representational Similarity Analysis (RSA).
I am currently using TDT to calculate the similarity (Spearman’s rho) of neural activation patterns across three conditions. To remove the influence of differences in the univariate responses between conditions, I am considering using scaling.

It seems common to use a scaling method called “cocktail-blank removal,” where the mean value of each voxel across conditions is subtracted. However, I don’t understand the purpose and necessity of using this scaling. Could someone explain this?
Additionally, if I set cfg.scale.method = ‘mean’ in TDT, is the scaling performed equivalent to cocktail-blank removal?

Thank you in advance for your help!